Boosting Alignment for Post-Unlearning Text-to-Image Generative Models

TL;DR

This paper introduces a novel unlearning framework for text-to-image generative models that effectively removes undesirable content while preserving model alignment, outperforming existing methods.

Contribution

The authors propose an optimal model update strategy for unlearning that ensures continuous improvement in both unlearning quality and text-image alignment.

Findings

Successfully removes target concepts from diffusion models

Maintains close alignment with original trained models

Outperforms state-of-the-art unlearning baselines

Abstract

Large-scale generative models have shown impressive image-generation capabilities, propelled by massive data. However, this often inadvertently leads to the generation of harmful or inappropriate content and raises copyright concerns. Driven by these concerns, machine unlearning has become crucial to effectively purge undesirable knowledge from models. While existing literature has studied various unlearning techniques, these often suffer from either poor unlearning quality or degradation in text-image alignment after unlearning, due to the competitive nature of these objectives. To address these challenges, we propose a framework that seeks an optimal model update at each unlearning iteration, ensuring monotonic improvement on both objectives. We further derive the characterization of such an update. In addition, we design procedures to strategically diversify the unlearning and remaining datasets to boost performance improvement. Our evaluation demonstrates that our method effectively removes target classes from recent diffusion-based generative models and concepts from stable diffusion models while maintaining close alignment with the models' original trained states, thus outperforming state-of-the-art baselines. Our code will be made available at https://github.com/reds-lab/Restricted_gradient_diversity_unlearning.git.